No abstract

All forecasts of Internet traffic point at a substantial growth over the next few years. From a network operator perspective, efficient in-network caching of data is and will be a key component in trying to cope with and profit from this increasing demand. One problem, however, is to evaluate the performance of different caching policies as the number of available data items as well as the distribution networks grows very large.In this work, we develop an analytical model of an aggregation access network receiving a continuous flow of requests from external clients. We provide exact analytical solutions for cache hit rates, data availability and more. This enables us to provide guidelines and rules of thumb for operators and Information-Centric Network designers.Finally, we apply our analytical results to a real VoD trace from a network operator and show that substantial bandwidth savings can be expected when using in-network caching in a realistic setting.

No abstract

The goal of this work is to develop efficient algorithms and data structures suitable for software-based forwarding of muRicast datagrams. The algorithms should scale to very large numbers of simultaneously active multicast groups. In an example configuration with 32 interfaces a new algorithm can forward 20,000 simultaneously active multicast groups using only 64Kbytes of memory and a leak probability of 2.4x10 -s. This algorithm can perform almost 10 million forwarding decisions per second on a 800MHz Penfium III processorThe task of a multicast-forwarding engine is to forward packets along branches of a distribution tree. The branches of the tree consists of network segments and the nodes in the tree are routers. There are a few different classes of distribution trees. For simpficity only bidirectional trees will be considered in this discussion.Abstractly, a bidirectional-distribetion tree is a bidirectional acyclic graph. R has a root called center, core or, rendez-vous point. Forwarding is performed by having traffic flow up the tree towards the center, and down the tree towards the receivers.A router must for each distribution tree maintain a list of interfaces that connect the tree. To forward a packet, first find the interface list corresponding to the group. If the packet arrived on an interface in the list, then, forward the packet over the remaining interfaces in the list.It is possible to perform this forwarding function using probabilistic methods. Observe that forwarding will be successfuUy performed if the interface list includes at least all interfaces that the packet must be forwarded over. Since the interface list can potentially be to long, a packet will occasionally leak out the wrong interface.However, at a minimum, packets will always be forwarded over the distribution tree. Since the leak probability can be made arbitrarily small, this may not be a problem in practice.Assume a network configuration where packets are allowed to leak with a maximum probability P. The probability that a leaked packet will leak through the next router is p2, and through a third p3. Thus, leak probabifity decays exponentially as we travel down (or up) the tree. Also, hosts configure their network adaptors to filter out unwanted multicasts. If a packet should sneak through this hardware filter, then software filtering sets in. Thus, hosts on leaf networks are already prepared to handle the problem with traffic leaks.The suggested probabilisfic-forwardin g algorithm has an efficient implementation based on Bloom filters. The algorithm uses a fixed small number of memory references (e.g eight) and a small number of multiplications and shift instructions. Memory requirements scale finearly with the maximum allowed number of groups and is also independent of the number of interfaces. Surprisingly, memory requirements are independent of group address length. This is a result of that addresses are not stored in the data structure. Thus, 112 bit IPv6 addresses use no more forwarding memory than 28 bit IPv4 addresses.A...